Hi, and thank you for forwarding the mails.

Note: gmail sends to both you and libav ml, so maybe an equivalent of
"reply all" may be needed to avoid private replies.

2016-04-29 9:52 GMT+02:00 Alexandra Hájková <[email protected]>:
> All the work is here:
> https://github.com/sasshka/libav/tree/get_bits0

More info from me:
stripped binary size goes from 10639872, for libav/master, to
11278336, ie around 600K. This makes me think that, whatever inlining
or reason for code size increase there is, the cause would better be
isolated, and ideally restricted to performance-sensitive code (other
parts ideally not using what causes that size increase, like there is,
iirc, av_cold).

The threshold could be low, maybe 600K is not something to care about.

Now that I have understood that not all codecs are yet converted in
the patchset, I have looked at a typical interesting case, huffyuv.
Example: -i <large sequence> -s 1920x1080 -an -sn -c:v ffvhuff
-vframes 500 -y out.avi

The results are interesting:
time /  size / decode_gray_bitstream _TIMER
4.16 / 34236 / 190887 UNITS in gray, 262143 runs, 1 skips
6.41 / 16080 / 283210 UNITS in gray, 262144 runs, 0 skips

I think this looks good and warrants attention (for those that care).

-- 
Christophe
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to